Hybrid Convolutional Neural Network Architectures for Skin Cancer Classification

Abstract

Skin cancer is a common form of cancer seen in humans. Like other diseases, early diagnosis of skin cancer is vital. In the study, deep learning architectures, which are popular machine learning algorithms, are used to classify skin cancer. In order to increase accuracy performance, hybrid structures are realized using K-Nearest neighbor (KNN), Support vector machine (SVM) and Decision tree (DT). After feature extraction using convolutional neural network, KNN, SVM and DT are applied separately for classification. While the KNN and SVM of the produced hybrid structures increase performance, the use of the decision tree has negatively affected the performance. After the training and validation processes with the seven-class skin cancer mnist: ham10000 dataset containing dermatological images, the validation accuracy and confusion matrix criteria of the architectures are reported. Eight different architectures are implemented. The highest accuracy is provided by the structure in which the last layer of Alexnet architecture is replaced by the SVM classifier.